The methods used by the international collaborative of macronutrients and blood pressure have been described in detail.26
Briefly, the study population (1996-9) comprises 2359 men and 2321 women aged 40-59 from 17 samples: Japan (four samples), the People’s Republic of China (n=3), the United Kingdom (n=2), and the United States (n=8). Data were collected cross sectionally. Each sample was representative of a defined target population; samples were included from both the community and the workforce. People were selected randomly from population lists, stratified by age and sex to give about equal numbers (65 people) in each of four 10 year age and sex groups.26
The mean participation rate was 49%. Each participant attended the research clinic on four occasions: two visits on consecutive days, a gap averaging three weeks, then two further consecutive visits.
Blood pressure was measured twice at each clinic visit with a random zero sphygmomanometer. Measurements were taken with the participant seated, after a rest of at least five minutes and with the bladder emptied. Korotkoff sounds I and V were used for systolic and diastolic blood pressures. At two visits height and weight were measured and questionnaire data were obtained on daily alcohol intake over the past seven days and on possible confounders.
At each visit a trained interviewer used the in-depth multipass 24 hour recall method to collect data on diet. This method uses open ended questions, is suited to multiple ethnicities and languages, does not require participants to be literate, and is designed to capture total dietary intake (macronutrients and micronutrients) for 24 hours.27
We therefore recorded all foods and drinks, including supplements, consumed in the previous 24 hours. Quality control measures were extensive.28
Participants gave written informed consent on their first clinic visit, before data collection.
We carried out two timed 24 hour urine collections, coinciding with the two pairs of consecutive clinic visits. Measurements taken at a central laboratory included sodium, potassium, and creatinine levels. To estimate technical error we split about 8% of local specimens before shipment to the central laboratory.26
We excluded those who did not attend all four clinic visits, those with dietary data considered unreliable, those whose energy intake from any 24 hour recall was less than 2.1 MJ or more than 21 MJ (women), or 33 MJ (men), those with two unavailable urine collections, and those whose data on other variables were incomplete or indicated violation of the protocol. Overall, we excluded 215 people. For each participant excluded we recruited another participant.
To provide comparable data on 83 nutrients across the four countries, including iron, we converted data on food and dietary supplements into nutrient intakes by using country specific food tables, updated and standardised by the Nutrition Coordinating Centre, University of Minnesota.29
We calculated nutrient intakes from only foods and from foods plus dietary supplements. For nutrients supplying energy we calculated intake as percentage millijoules and for others as intake per 4.2 MJ. We also calculated nutrients as amount per 24 hours. To estimate intake of haem and non-haem iron from dietary total iron intake we used an established formula,30
taking into account the source of iron (meat, poultry, fish v
other). We used the data on food to characterise the main food groups supplying iron and to estimate each participant’s red meat intake (defined as skeletal muscle tissue of all mammals) and their consumption of beef. Measurements for each participant were averaged for blood pressure and nutrient variables across the four visits, and for urinary excretions across the two 24 hour collections.
From the mean of the four visits we estimated the reliability of iron and red meat intake for individual participants using the formula 1/[1+(ratio/4)]×100, where the ratio is the variance within a participant divided by the variance between participants, calculated separately for eight sex and country strata and pooled by weighting of each stratum specific estimate by n−1. This gives an estimate of the size of an observed regression coefficient as a percentage of the theoretical coefficient in a univariate regression analysis of blood pressure on iron, red meat, or beef.31 32
We used partial correlation to explore the associations among dietary variables, adjusted for sample, age, and sex, and pooled across countries and weighted by sample size. To examine the relations of iron intake (total, haem, non-haem, total plus supplements, all as mg/4.2 MJ) and intake of red meat or beef (grams over 24 hours, adjusted for energy intake) to systolic and diastolic blood pressure we used multiple linear regression analyses. Adjustment for possible confounders was done sequentially: for sample, age, sex, weight, height, reported special diet, use of dietary supplements, moderate or heavy physical activity (hours daily), doctor diagnosed cardiovascular disease and diabetes, and family history of hypertension (model 1); plus 24 hour urinary excretion of sodium and potassium (or urinary sodium or creatinine, potassium to creatinine ratios) and alcohol intake over 14 days (model 2); plus levels of dietary cholesterol, saturated fatty acids, and polyunsaturated fatty acids (model 3); plus intake of either animal protein, vegetable protein, dietary fibre, magnesium, phosphorus, calcium, haem iron (red meat analysis only) or non-haem iron (red meat analysis only) entered separately to avoid colinearity (models 4a-h).
To estimate the overall association we fitted regression models by country and coefficients pooled across countries, weighted by the inverse of variance. We tested heterogeneity across countries and we assessed interactions for age and sex. Because vitamin C and alcohol increase the intestinal absorption of non-haem iron we tested interactions with these two variables in models examining non-haem iron.22 23
We carried out regression analyses on four cohorts: all 4680 participants; 2238 “non-intervened” participants (not receiving a special diet; not consuming dietary supplements; no diagnosed cardiovascular disease and diabetes; not taking drugs for high blood pressure, cardiovascular disease, or diabetes—that is, exclusion of people whose characteristics might bias the relations between iron and blood pressure10 11
); 3671 non-hypertensive participants (systolic blood pressure <140 mm Hg, diastolic blood pressure <90 mm Hg, not taking antihypertensive drugs); and 2038 “non-intervened” non-hypertensive participants.
Sensitivity analyses concerned the analysis of iron from food plus dietary supplements (with and without exclusion of 15 men and 44 women with a supplemental iron intake >30 mg/day, indicative of chronic iron deficiency); adjustment for total carbohydrate, starch, or total sugars (percentage millijoules); use of nutrient densities of iron (haem, non-haem, and total) adjusted for energy; use of amount per 24 hours of iron intakes (haem, non-haem, and total) adjusted for energy; and exclusion of people (predefined) with noticeable variability in nutrient intake or blood pressure. For ease of interpretation we transformed regression coefficients to represent difference in blood pressure for each two standard deviation higher intake of iron (haem, non-haem, or total), red meat, or beef. We adjusted each standard deviation for country, estimated by the mean square error from country adjusted analysis of variance.
Analyses (IT, IJB) were done with SAS version 9.1. We considered a P value <0.01 to be significant. A P value <0.05 was used for interaction terms and tests for heterogeneity across countries.